C++ Concurrency in Action

Synchronization proofs are more reliable than timing guesses

First edition chapter 2, 2.1–2.3; chapter 4, 4.1; chapter 5, 5.2–5.3; chapter 7, 7.2; chapter 10, 10.2

Whether a thread has finished, whether state is readable, and whether an object is still alive are three different questions. The first edition writes from thread ownership through shared state, condition variables, and memory order, offering the common perspective of looking for an explainable happens-before relationship rather than trusting that the machine is slow enough or that one has waited long enough. Notification itself does not store business facts; the predicate protected by the mutex stores the facts. Likewise, each update of an atomic counter is indivisible, but that does not automatically make a multi-field protocol hold. Reading the lock-free chapters also requires separately proving node reclamation and pointer updates; a successful compare-exchange cannot substitute for object-lifetime analysis. For today's engineering I preferentially keep easy-to-review locks and explicit lifetimes; only when measurement proves a bottleneck exists do I discuss splitting locks or lowering memory order. C++20's jthread shortens the boilerplate for reclaiming threads, but it does not design the shutdown protocol for us.